Defaults at a 13-year high, no recession

I watched it happen. A friend left his salaried job last spring and bought a small heating and air company one town over.

He put down a tenth of the purchase price. A bank lent him the rest.

Then the federal government guaranteed most of what the bank lent. That is how a great many small businesses in America change hands.

The program is called 7(a). It has run since 1953. By law it is supposed to cost the taxpayer nothing… lender fees pay for the losses.

For most of the last decade that held.

As of March 31 this year, the twelve-month default rate across the whole 7(a) book reached 4.8%. That is the highest reading since 2013.

But the level is not the strange part. The timing is.

Default rates climb in recessions. That is what they do. In 2002 this one touched 5.1% as the dotcom bust worked through the economy. In 2010 it reached 11.6%, in the worst credit event since the Depression.

There is no recession now. Unemployment is low. The stock market sits near its highs. Earnings are growing.

And this book sits within half a point of its dotcom peak.

So who exactly is defaulting?

This portfolio is built, by statute, out of the borrowers a bank will not take on its own terms.

The 7(a) program carries a rule called the credit elsewhere test. A business qualifies only if it cannot get a loan on reasonable terms without the guarantee.

So the portfolio is a portrait of the marginal small business owner. The one already sitting at the edge of what a lender will fund.

When conditions tighten, those owners move first. Everyone with room absorbs it for a while longer.

Meanwhile the headline numbers look calm. Business loan delinquency at all commercial banks ran 1.3% in the first quarter. Wells Fargo's commercial charge-off rate fell to 0.1% in the second.

Both readings are correct. They describe different people.

I have no idea when this turns into something larger. It may not.

But a loan book reaching its second-worst level in thirty years without a recession is worth a second look. That's interesting.

This connects directly to the borrower question. If a whole economy were softening, old loans and new loans would sour together. They are not.

Line up every class of SBA 7(a) loans by age instead of by calendar year and the picture separates cleanly.

In their first full year on the books, loans from fiscal 2024 and fiscal 2025 defaulted at 3.4% and 3.5%. The 2016 class defaulted at 1.6% at the same age.

By year two the gap widens. The 2024 class is at 7.2%. The 2016 class was at 3.0%. By year three, the 2023 class is at 8.0% against 3.1%.

A loan that fails inside eighteen months was not overtaken by events. Rates did not move that far, that fast, for that one borrower. Lenders call it early default, and it measures the decision rather than the weather.

So these loans were misjudged at the desk. Not undone by the economy.

And here's where it spreads. One deteriorating loan programme proves very little on its own. So look at the instruments that only capture small firms.

Subchapter V is the restructuring path Congress reserved for small businesses. Filings hit 1,663 in the first half of 2026, against 1,107 in the same months of 2025.

The count should have gone the other way. In mid-2024 the temporary eligibility ceiling of $7.5 million lapsed, which removed a whole band of larger filers from the path. Fewer firms can use it now. Half as many again are using it.

The Federal Reserve's own survey of employer firms says the same thing in a softer voice. Expectations for revenue and employment growth have fallen to their lowest since 2020. For the second survey running, more firms reported revenue falling than rising.

More than four in ten named tariff costs as a challenge. Thirty-eight per cent carry more than $100,000 of debt.

And in January the Fed asked banks what they expected across this year. Loan quality would hold up for large and middle-market borrowers, they said. It would get worse for small firms.

Meanwhile, the shape of the risk changed underneath everyone.

In 2016 the ranking of SBA 7(a) loans by size was a tidy ladder. Loans under $150,000 defaulted at 2.5%, the worst band. Loans above $3 million defaulted at 1.1%, the best. Bigger loan, better borrower.

In 2025 the ladder became a U. The smallest loans are still worst at 4.8%. But the $3 million bucket now sits at 4.4%, third from the top. The safest place in the book is the unglamorous middle, at 3.3%.

Consider what a $3 million loan of this kind actually buys. At that size it is almost always somebody purchasing a business outright, or buying out a partner. Heavy on goodwill, thin on slack, and floating against prime.

Debt service on four million dollars does not travel gracefully from 6% to 10.5%.

Among the 36 largest holders, one lender ran a 0.5% default rate last year. Another ran 12.3%.

Both operate at scale. Both lend under the same rulebook, with the same federal guarantee, into the same economy. Eleven of those 36 are running below their own 2019 default rate.

So this is not only a story about conditions. A good deal of it is a story about who was writing the paper.

Three more things worth keeping track of…

Japan gave two answers to the same question last month. The TOPIX finished July broadly flat. The Nikkei 225 fell almost 8%. Same country, same four weeks. The gap is sector weight, nothing else: the Nikkei leans on technology, while the TOPIX carries more banks, industrials and domestic businesses. When one nation's two headline indexes disagree by eight points in a month, the index is telling you about its own construction rather than about Japan.

Semiconductors had a violent July that the calendar year hides. The Philadelphia Semiconductor Index fell 20.6% over the month, its worst since 2008, and still ended up close to 60% for the year. SK Hynix lost 35% and Samsung Electronics lost 21% across the same weeks. A drop that size inside a gain that size means the year's return rests on a very narrow base.

And American households are pre-paying for power. PJM runs the grid for thirteen states and Washington, D.C. Its latest capacity auction, covering June 2028 through May 2029, produced $16.4 billion of capacity charges. About $6.3 billion of that traces to data-centre demand, according to Monitoring Analytics, the grid's own independent market monitor. Across the last four auctions the figure comes to $29.4 billion. The average residential rate is now 18.44 cents per kilowatt hour, up 6.2% on the year. People are paying today for plants that do not exist yet, to serve customers who have not arrived yet. We'll see.

In 1968 Congress passed the Housing and Urban Development Act. It contained a programme called Section 235.

The idea was a decent one. A low-income family could buy a house with $200 down. Some paid as little as 1% interest. The Federal Housing Administration insured the mortgage.

That last part changed everything.

If the family stopped paying, the lender collected in full from the government anyway. So the condition of the house stopped mattering to the person lending the money.

Speculators worked out the arithmetic fast. Buy a run-down house cheap. Paint it. Find an appraiser who will sign a number. Congressional investigators later found federal appraisers taking bribes and valuing houses at three and four times what they were worth.

The families moved in and found leaking roofs and furnaces that did not work. They could not afford the repairs. So they left.

By the end of 1973, one in ten Section 235 homes sat in foreclosure. In May 1974 the government owned 78,000 foreclosed single-family houses.

A guarantee does not remove a loss. It only decides who gets to ignore it, and for how long.

By 1974, twenty-eight officials at Housing and Urban Development had been indicted. The FBI was running 1,930 open fraud investigations. Nixon halted the programme in 1973, and Congress finally ended it in 1987.

Here is the part I keep coming back to. This was not a fringe experiment running off in a corner.

In April 1970 the research director of the Mortgage Bankers Association put Section 235 loans at three-quarters of most mortgage bankers' business. The programme was the industry.

In other words, an entire lending trade reorganised itself around a promise that somebody else would take the loss. And it took six years and 78,000 empty houses for anyone to add up the bill.

A federal guarantee never made a bad loan good. It only moved the loss somewhere nobody had to look at for a few years.

We'll see.